Unigine Heaven Benchmark with DX11 Tessellation

Brent_Justice

Moderator
Joined
Apr 17, 2000
Messages
17,755
Unigine Heaven Benchmark with DX11 Tessellation -
What does DX11 Tessellation look like in a real-time 3D engine? Unigine’s Heaven Benchmark supports DX11 with Tessellation and we’ve taken a look at Tessellation in this unique application. Tessellation has the potential to improve the gameplay experience, and you’ll see why once you see these screenshots.

BTW, I know people will want to compare their results with ours, being a benchmark and all. Just launch the program and set the resolution to 1680x1050 Full Screen, and make sure shaders are on High, leave AA and AF at defaults. On DX10 level cards use the DX10 render, on DX11 cards you can use DX10 and DX11 render.
 
Last edited:
I mean in this regard developers can make use of a feature on the graphics card, That doesn't mean that games will use it.

Some places where people use the ingame engine for cut-scene's that springs to mind as something that can have some extra detail without it hurting performance.

And these days the average development time for a game is longer and longer, this means if were getting games that use these kind of features that it might be a while,

Programmers can/will tweak their polygon models for these kind of functions? Games can end-up looking piss poor on cards that do not support hardware tessellation.
 
do keep in mind that in the comparison shots, in DX10 they could have added instanced tessellation which is supported by the older ATI/NVIDIA hardware; but they did not, making the on/off comparison a bit exaggerated.

Instanced Tessellation in DirectX10

if they'd added some instanced tessellation to the DX10 path the difference would be noticeably less stellar!
 
I mean in this regard developers can make use of a feature on the graphics card, That doesn't mean that games will use it.

Some places where people use the ingame engine for cut-scene's that springs to mind as something that can have some extra detail without it hurting performance.

And these days the average development time for a game is longer and longer, this means if were getting games that use these kind of features that it might be a while,

Programmers can/will tweak their polygon models for these kind of functions? Games can end-up looking piss poor on cards that do not support hardware tessellation.

Fact is we know of two games right now that will use DX11 Tessellation, DiRT 2 will use it for the water and cloth and crowd. Here is a screenshot of that wireframe - http://hardocp.com/image.html?image=MTI1MzU4OTM1NVlDbXBla3ZKZm5fNV8xOV9sLmdpZg== and there is a DX11 Tech video out there as well here - http://www.youtube.com/watch?v=D9p3PYOX1Vc

Alien vs. Predator will use it on the Aliens, here is a wireframe screenshot of that - http://hardocp.com/image.html?image=MTI1MzU4OTM1NVlDbXBla3ZKZm5fNV8yMF9sLmdpZg== - Not as detailed as the Dragon in this benchmark, but the ability is there in the engine itself.
 
I must say that ever since I first saw this demo something has bothered me. Is it just me or does that road look extremely uncomfortable? It's all jagged with sharp pointy rocks......:(
 
Tell me a game that doesn't look piss poor when you lower the graphics settings?
 
do keep in mind that in the comparison shots, in DX10 they could have added instanced tessellation which is supported by the older ATI/NVIDIA hardware; but they did not, making the on/off comparison a bit exaggerated.

Instanced Tessellation in DirectX10

if they'd added some instanced tessellation to the DX10 path the difference would be noticeably less stellar!

Fact is, these features in games, such as Dirt 2 and Alien vs. Predators, are coming under the DX11 API for DX11 Tessellation. We won't see DX10 Instanced anything in games. We indicated in the article that Tessellation is nothing new, the idea certainly isn't, but the ability to run it on consumer hardware at playable framerates using the latest API sure is. The fact it is coming in games makes it relevant now.
 
I mean in this regard developers can make use of a feature on the graphics card, That doesn't mean that games will use it.

Some places where people use the ingame engine for cut-scene's that springs to mind as something that can have some extra detail without it hurting performance.

And these days the average development time for a game is longer and longer, this means if were getting games that use these kind of features that it might be a while,

Programmers can/will tweak their polygon models for these kind of functions? Games can end-up looking piss poor on cards that do not support hardware tessellation.

I fully expect this to make it into most games, as from what I understand it is a relatively easy thing to enable that requires minimal developer intervention. If anything, I expect developers to use it as a crutch. Port over the models from the console versions, and then use tessellation to add details dynamically instead of creating and tweaking higher poly models manually.

Time will tell, of course :)
 
If anything, I expect developers to use it as a crutch. Port over the models from the console versions, and then use tessellation to add details dynamically instead of creating and tweaking higher poly models manually.

But at the least, it's still gonna be better graphically than if it was straight ported. Either way, dynamic or manual, we get better graphics. :cool:
 
Isn't it actually displacement mapping? at least on the road and the dragon.
 
From the images it looks like the Tessellation is operating on a bump map, it should not be too difficult to implement.
 
I must say that ever since I first saw this demo something has bothered me. Is it just me or does that road look extremely uncomfortable? It's all jagged with sharp pointy rocks......:(

It's over exaggerated that's for sure. But I think once it is scaled down to a reasonable level this will look AMAZING. This is the kind of "pop" that DX10 vs DX9 just didn't have in most games. When a larger selection of games start hitting the market with this I think I'll finally have a reason to upgrade.
 
But how does the dyanmic method determine how it should be shaped? What should be squared, pointy, or sharp? The dragons pointed scales COULD be rounded.

How is tessellation used currently that it isn't dynamic?
 
My concern is that in the last two full size screenshots the house in the rear has no tessellation applied to it.

I understand this is a minor point but it really breaks the immersion factor of an otherwise awesome technology.

I also hope it is applied more realistically in future titles. Time will tell.

I have high hopes for this technology.
 
I must say that ever since I first saw this demo something has bothered me. Is it just me or does that road look extremely uncomfortable? It's all jagged with sharp pointy rocks......:(

I was thinking the same thing. ;) It would suck to walk on lmao.

But this is very cool stuff. Aliens in the new AvP look absolutely badass with tessellation. Can't wait for the [H] review on this tech in games to see if it will finally be a great benefit to games. Honestly I loved the hardware tessellation my Radeon 8500 (or was it the 9700?) back in the day with Counter-Strike. Sure it made some guns look a little funky but overall it was neat tech back then but being proprietary at the time it wasn't excepted and shouldn't have been. Now that it's back in an open standard I say bring it on!
 
My concern is that in the last two full size screenshots the house in the rear has no tessellation applied to it.

I understand this is a minor point but it really breaks the immersion factor of an otherwise awesome technology.

I also hope it is applied more realistically in future titles. Time will tell.

I have high hopes for this technology.

I hadn't even noticed that. I don't find it a break in the immersion factor except for the fact that house will suddenly "explode" as tessellation is applied as you get closer. :eek:
 
i thought there are games that uses this type of feature before. Doesn't unreal 3 engine do something similar? Loads different levels of detail over when you get closer?
 
I know you guys don't have the resources to test out every possible configuration of video cards, but do you think tessellation will see a benefit from multiple GPU setups? With the load balanced between two 5870s, do you think frame rates would see an appreciable jump?
 
I know you guys don't have the resources to test out every possible configuration of video cards, but do you think tessellation will see a benefit from multiple GPU setups? With the load balanced between two 5870s, do you think frame rates would see an appreciable jump?

I'll second this. Lets see some multi-GPU numbers! I think it's pretty clear that the benchmark needs it!

I'm guessing there isn't a crossfire profile for the benchmark, but I believe there is a tool to create one IIRC. So that shouldn't stop you. However, I'm wonder if you guys have a second 5870 I know they are in high demand atm.
 
Basically Brent it is not about the 2 games. It is about the whole thing where developers are either able to make good use of it or just don't bother at all.

In the Unigine engine it is a given that this feature can used easily. Not every framework for these engine's have the same quality nor the same feature's to say that their engine will sell because of hardware tessellation is another thing.

If you look at some MMORPG you can see that there not all using the same engine. Goes for other games as well.

Where you expect developers to jump on a certain bandwagon is also limited by hardware support/sales.

This is where i frown upon some things, higher poly models for "free" okay. But still if it was that easy to be implemented then everyone already done it :) .
 
Basically Brent it is not about the 2 games. It is about the whole thing where developers are either able to make good use of it or just don't bother at all.

In the Unigine engine it is a given that this feature can used easily. Not every framework for these engine's have the same quality nor the same feature's to say that their engine will sell because of hardware tessellation is another thing.

If you look at some MMORPG you can see that there not all using the same engine. Goes for other games as well.

Where you expect developers to jump on a certain bandwagon is also limited by hardware support/sales.

This is where i frown upon some things, higher poly models for "free" okay. But still if it was that easy to be implemented then everyone already done it :) .

Oooook. Not sure what your point is, two games (one coming out this December and the next early next year) are going to utilize this feature, and there are more in the pipeline. So, apparently this technology is going to be used via DX11 in published games. What I've seen in Dirt 2 and AvP seems to look like good use of it to me, but we'll certainly evaluate that when the games are released.
 
It was mentioned in the article, but I think people and developers would feel a bit better about Tessellation if there was a way to adjust the amount of tessellation going on in the Heaven benchmark. As it is, the volume of it looks massive, overdone. I'd like to see it turned down a little, to see how much of it is necessary to change the look of what without it is a flat texture into something more living (long story, but it is so much different from displacement mapping) and how much performance you'll lose out on.

(I certainly hope they give that freedom in the games)

Tessellation so far has been demonstrated as a binary, either it's off, or it's on and things have a lot more detail and the framerate takes a big hit - I want to see what's inbetween.

(Very good article as always. Looking forward to seeing the NDA'd tessellation using games)
 
As it is, the volume of it looks massive, overdone. I'd like to see it turned down a little, to see how much of it is necessary to change the look of what without it is a flat texture into something more living (long story, but it is so much different from displacement mapping) and how much performance you'll lose out on.

The look reminds me of truform on serious sam
 
Tessellation so far has been demonstrated as a binary
Not true, actually. I definitely recall a demo that showed a somewhat cartoony looking head. As the demonstrator increased the tessellation slider, the model smoothed out quite a bit. It didn't really add extra details per se, it just refined the higher-order primitives in the model, but it definitely showed user-control of tessellation in real time. I'll try to find a link for you.
 
Wasn't tessellation available in ATI video careds since the HD2000 series came out? The HD2900 had a Programmable tessellation unit. What I'm wondering now is how where developers supposed to take advantage of it, because at that time, the HD2000 series was, DirectX 10, and tessellation has only now been implemented in DirectX 11. Also, why did ATI put effort and resources back then into implementing a technology at a hardware level, that almost no one could take advantage of. I'm not saying it as a bad thing, but gamers that still own a HD2900 won't be able to take advantage of it, because their cards don't support DirectX 11. It's kind of a bummer. It seems to me that NVidia is only implementing as much as necessary with every graphics card generation, and doesn't bother with point releases of DirectX either, and as a result, their cards seem to be more focused on what matters. I'm only saying this in reference to the HD2900 card, not to the current generation of ATI cards. The HD2900 had amazing specs on paper, but it was a crappy performer because it lacked in other areas like anti aliasing. Anyway, my whole point is that this technology has been available since the HD2000 series cards (in ATI cards), and I don't think anyone that owns a HD2000, HD3000 or HD4000 series card can take advantage of it, just folks with HD5000 cards. So what was the point of implementing it way back then?
 
I'll second this. Lets see some multi-GPU numbers! I think it's pretty clear that the benchmark needs it!

I'm guessing there isn't a crossfire profile for the benchmark, but I believe there is a tool to create one IIRC. So that shouldn't stop you. However, I'm wonder if you guys have a second 5870 I know they are in high demand atm.

I have (2) XFX 5870's in Crossfire and I am willing to run any tests/settings you guys would like.

Here is a post I made in the 9 page Heaven Benchmark thread: (Tessellation was disabled as I was matching settings with a fellow poster who had a nVidia card)

http://hardforums.com/showthread.php?t=1462534&page=9

only thing that has changed from my 124 score is my GPU core(s) has been bumped to: 900mhz and my GPU memory clock is at 1300mhz
 
Wasn't tessellation available in ATI video careds since the HD2000 series came out? The HD2900 had a Programmable tessellation unit. What I'm wondering now is how where developers supposed to take advantage of it, because at that time, the HD2000 series was, DirectX 10, and tessellation has only now been implemented in DirectX 11. Also, why did ATI put effort and resources back then into implementing a technology at a hardware level, that almost no one could take advantage of. I'm not saying it as a bad thing, but gamers that still own a HD2900 won't be able to take advantage of it, because their cards don't support DirectX 11. It's kind of a bummer. It seems to me that NVidia is only implementing as much as necessary with every graphics card generation, and doesn't bother with point releases of DirectX either, and as a result, their cards seem to be more focused on what matters. I'm only saying this in reference to the HD2900 card, not to the current generation of ATI cards. The HD2900 had amazing specs on paper, but it was a crappy performer because it lacked in other areas like anti aliasing. Anyway, my whole point is that this technology has been available since the HD2000 series cards (in ATI cards), and I don't think anyone that owns a HD2000, HD3000 or HD4000 series card can take advantage of it, just folks with HD5000 cards. So what was the point of implementing it way back then?

nicepoint it would be better if there was some kinda of patch for post 2000 series users to use tesellation thing without it that feature seems really useless for pre 5000 series card

I have (2) XFX 5870's in Crossfire and I am willing to run any tests/settings you guys would like.

Here is a post I made in the 9 page Heaven Benchmark thread: (Tessellation was disabled as I was matching settings with a fellow poster who had a nVidia card)

http://hardforums.com/showthread.php?t=1462534&page=9

seems this engine loves cf amazing fps Hemlock gonna be killer in tesellation used games
 
The "DX10" version looks more like DX7 at min settings with high res textures - I mean without tessellation the stairs are so low poly they are a flat slope (i.e. no stairs).

I also agree with others that you could achieve something with 95% of the look and feel and a low poly count just using DX9c and displacement mapping - remember the UE3 promotional stuff that said the displacement mapping made things look like they have 100 times the amount of polys they actually have.

I also agree you need to keep the detail level fairly consistent - so having really flat looking very low poly bits interspersed with uber high detail tessellated bits looks fake.

In many ways tessellation boils down to an auto lod-ing method (lod = level of detail), where the lod is generated by the graphics card as opposed to the progamme manually switching to a different set of geometry - but the effect is the same (i.e. the moment you get X distance away from object X switch to a higher polygon count version). Any DX version of a game could use lod's.

Hence the real question is what method gives the best performance. i.e. unlike this demo we want a demo where there is only one graphics quality - very high, but you can use different methods to achieve that - e.g. tessellation,lods,highest detail all the time, medium detail + displacement maps (to give the same high detail). The difference should be in the fps you get (I'd hope tessellation came out top or there would be no point using it).
 
I do not see the big deal. You can have triple the number of polygons and it still isnt visually "compelling".
Realistic lighting is much more compelling visually than more polygons. We have been doing the "more polygons" thing for 10 years now and games still look the same. Take an old game with 2002 era polygon counts and give it fully realistic lighting and it will be much more impressive than anything in these screen shots....
 
I have (2) XFX 5870's in Crossfire and I am willing to run any tests/settings you guys would like.

Here is a post I made in the 9 page Heaven Benchmark thread: (Tessellation was disabled as I was matching settings with a fellow poster who had a nVidia card)

http://hardforums.com/showthread.php?t=1462534&page=9

Just looking for scaling preformance in crossfire at an enthusiast (i.e. 1920x1200 or 2560x1600) with tesolation.
 
Obviously a developer might not want to use this much in-depth Tessellation since it causes a large performance drop.
The only question I have is how the Programmers will optimize their tessellation models to gain more performance. I don't mean less polygons but more polygons only where they're needed.
For example:
http://hardocp.com/image.html?image=MTI1NzM4MjE1MjZpek04cDRMQWxfMV8xM19sLnBuZw==

There are so many polygons used to create that round vent and give it a 3D appearance when at first sight you might not even notice the difference with the non tessellated one. If you know the shape you want to draw there's simpler geometry to obtain a very similar result, resulting in less polygons.

Another example:

http://hardocp.com/image.html?image=MTI1NzM4MjE1MjZpek04cDRMQWxfMV84X2wucG5n

This was already commented when the first video of Unengine was posted here.

Who would want to walk on that road?
This is clearly and exaggeration and by the number of polygons it takes to build a road that is unbelievably bumpy. That geometry could be used to a lesser extent with the same purpose.
The stairs next to it look quite great on the other hand and the polygon count seems fair.

Last:
http://hardocp.com/image.html?image=MTI1NzM4MjE1MjZpek04cDRMQWxfMV8xNF9sLnBuZw==
You can clearly see the difference in the window's roundness but the ammount of polygons it takes to make such difference is astonishing.
 
The "DX10" version looks more like DX7 at min settings with high res textures - I mean without tessellation the stairs are so low poly they are a flat slope (i.e. no stairs).

I also agree with others that you could achieve something with 95% of the look and feel and a low poly count just using DX9c and displacement mapping - remember the UE3 promotional stuff that said the displacement mapping made things look like they have 100 times the amount of polys they actually have.

I also agree you need to keep the detail level fairly consistent - so having really flat looking very low poly bits interspersed with uber high detail tessellated bits looks fake.

In many ways tessellation boils down to an auto lod-ing method (lod = level of detail), where the lod is generated by the graphics card as opposed to the progamme manually switching to a different set of geometry - but the effect is the same (i.e. the moment you get X distance away from object X switch to a higher polygon count version). Any DX version of a game could use lod's.

Hence the real question is what method gives the best performance. i.e. unlike this demo we want a demo where there is only one graphics quality - very high, but you can use different methods to achieve that - e.g. tessellation,lods,highest detail all the time, medium detail + displacement maps (to give the same high detail). The difference should be in the fps you get (I'd hope tessellation came out top or there would be no point using it).


What about the rope? Have you seen that rope?
 
Just ran benchmark same settings as [H]ard OCP but I have two Sapphire 5870 in Xfire and my score was 74.2. Will be interested in others scores with Xfire,
 
Wasn't tessellation available in ATI video careds since the HD2000 series came out? The HD2900 had a Programmable tessellation unit. What I'm wondering now is how where developers supposed to take advantage of it, because at that time, the HD2000 series was, DirectX 10, and tessellation has only now been implemented in DirectX 11. Also, why did ATI put effort and resources back then into implementing a technology at a hardware level, that almost no one could take advantage of. I'm not saying it as a bad thing, but gamers that still own a HD2900 won't be able to take advantage of it, because their cards don't support DirectX 11. It's kind of a bummer. It seems to me that NVidia is only implementing as much as necessary with every graphics card generation, and doesn't bother with point releases of DirectX either, and as a result, their cards seem to be more focused on what matters. I'm only saying this in reference to the HD2900 card, not to the current generation of ATI cards. The HD2900 had amazing specs on paper, but it was a crappy performer because it lacked in other areas like anti aliasing. Anyway, my whole point is that this technology has been available since the HD2000 series cards (in ATI cards), and I don't think anyone that owns a HD2000, HD3000 or HD4000 series card can take advantage of it, just folks with HD5000 cards. So what was the point of implementing it way back then?

You can blame microsoft for that.All this stuff was promised in DX10.ATI showed up ready NV didn't and Microsoft got retarded
 
I find that the engine shows tessellation in a far to extreme way..

Does anyone here remember the Farcry upgrade patch? It feels like that right now... It looks all fancy, but when shown in game its all smoke..Or the hilariously crazy Flightsim dx9 vs dx10 screen shots.

The full screen shot of the town really bugs me.. If you look at the 'steps' on the bottom right its a ramp in the tessellation off and then steps on the tessellation on... This to me looks like they are trying REALLY hard to fake the need for it.. Why wouldn't that have been steps in the first place?

I'd prefer some better lighting and affects.. I find the polygon count of a game does not add anything to the 'experience' for me.. Its the texture detail.. Where did the 64bit color textures go?? - when I walk up to a wall and its this blurry mess, it knocks me out of the game quick..

the next question - how much different would be performance be if they made the same demo with the same polygon count as the tessellation feature? That would mean DX10 hardware could run it, right? This of course would include the extra hours needed to manually interpret the object.
 
Just ran benchmark same settings as [H]ard OCP but I have two Sapphire 5870 in Xfire and my score was 74.2. Will be interested in others scores with Xfire,

Of course, I do not know if this program utilizes CrossFire or SLI well, it may not benefit much from multi-gpu in its current state. I don't really know, so we left out that testing since this is just a benchmark afterall, not a game.
 
http://hardocp.com/image.html?image=MTI1NzM4MjE1MjZpek04cDRMQWxfMV8xNF9sLnBuZw==
You can clearly see the difference in the window's roundness but the ammount of polygons it takes to make such difference is astonishing.

Remember all it boils down to is having different levels of detail for objects. So in the demo you can clearly see in the distance the level of detail is low, as you get closer it swaps to a higher level of detail (using tesselation to provide the higher lod). You could just as easily achieve that by having the programme switching the geometry for the object to a higher level of detail one. Hence tessellation gives us nothing new that way - lods have been around since before DX was invented.

So what advantage does tessellation give us? - it can only be performance, but as this demo doesn't have a "DX10 with lods" mode we can't actually compare that.
 
What about the rope? Have you seen that rope?

Don't think you quite got what I meant?

For example you could achieve the pretty looking rope by:
a) using tessellation so when you get close to it, the graphics card generates a higher detail version.
b) using two versions of the geometry so when you get close to it the programme just switches to the version with lots of polys (identical in every way to the tessellated one).
c) using a displacement map - which done right would also look near identical.

Obviously (a) required DX11, you could do (b) with DX7, and (c) with DX9c. The only reason I can see to use (a) would be if it did it with less of a performance hit then the other ways? Unfortunately the demo doesn't help prove or disprove that.
 
I'm actually feeling less than impressed with the "dynamic" nature of this. How would a game dev control how dynamic tesselation looks??

Look the the places where I saw large differences between the dx10 and dx11 screenshots:
The road
The roof shingles
The dragon and its spiked scales

The road looks unnaturally uneven. Who would want that particular look? The extra detail seems a good idea, but dynamically decided by the engine? Why not just program in extra polys?

The roof shingles look alot better in the dx11 shot, but this is obviously NOT dynamic tesselation, they have a whole new shape that I really doubt is "dynamic". It's simply extra polys.

The Dragon, same as the roof. That's obviously not dynamic. One has the spikes, one doesn't have them at all. Perhaps I'm not understanding what this dynamic tesselation actually does?

So it appears to me that this demo was made to look purposely better in Dx11. Are these guys trying to sell 3d engines? Everything seen in the dx11 shots that "looks good" could be done just the same in dx10 with some extra polygons...

In any event, it seems no hardware can really run that many polys at 1920x1200 except possibly the top end 5870, and 22 frames is a bit on the low side. Its more likely someone would turn down some graphics options to get better performance.

Maybe in a year we will have powerful enough hardware to see it used more.

Can [H] show some other differences between dx10 and dx11?
 
Back
Top